Page 170 - Handbook of Biomechatronics

Biomechatronic Applications of Brain-Computer Interfaces 167

performance. For example, studies that experimentally related BCI classifi-
cation accuracy to user satisfaction by artificially inducing classification
errors have found that the relationship is highly nonlinear and occasionally
nonmonotonic (van de Laar et al., 2013; McCrea et al., 2017). Furthermore,
studies in related fields such as EMG-controlled prosthetics have found that
offline classification accuracy does not necessarily correspond to online
accuracy, as users will learn to compensate for systematic classification errors
and reduce their effect ( Jiang et al., 2014; Hargrove et al., 2010).
If possible, BCIs should not only be evaluated with regard to their func-
tional effect (communication speed, enjoyment, rehabilitation outcome,
wheelchair navigation speed, etc.), but should also be compared to other
control methods that could potentially achieve a better outcome or achieve
the same outcome more easily. For example, as SSVEP-based BCIs essen-
tially measure the focus of the user’s gaze, their performance could be com-
pared to that of an eye tracker, which measures gaze without the need to
attach electrodes to the head. Similarly, EEG-based difficulty adaptation
methods could be compared to performance-based adaptation methods,
manual adaptation by the user (though this is not recommended by some
researchers (Ewing et al., 2016)), or to simple random adaptation. Following
a performance analysis, additional cost-benefit analyses could be done to
qualitatively or quantitatively compare the different control methods with
regard to other factors such as setup time and required user training time.
In this way, the potential advantages and disadvantages of BCIs as well as
their suitability for different applications could be clearly defined, setting
the stage for real-world adoption.

3.5 Outlook
State-of-the-art BCIs have already proven their worth in several assistive
biomechatronic systems, and are regularly used by people with severe dis-
abilities who would otherwise not be able to perform everyday activities
or even communicate with their loved ones. Furthermore, through the
introduction of ERPs into the human-machine interaction process, they
are driving the development of a new generation of co-adaptive
biomechatronic systems that adapt to the user’s preferences, dislikes, and
mistakes. While the benefits of BCIs in some applications (e.g., difficulty
adaptation) are not yet clear, advances in hardware and software are rapidly
increasing both the performance and user friendliness of BCIs, which will
undoubtedly lead to their broader adoption in a number of fields.